Verlato G, Muggeo M
Medical Statistics, University of Verona Medical School, Italy.
Diabetes Care. 2000 Jun;23(6):759-64. doi: 10.2337/diacare.23.6.759.
The present investigation used data from the Verona Diabetes Study to verify a main assumption of the capture-recapture method (source independence) and to characterize the subgroup of known diabetic patients missed by all sources whose number is estimated by the capture-recapture method.
The Verona Diabetes Study identified 7,148 type 2 diabetic patients on 31 December 1986 using 3 different sources: family physicians, a diabetes center, and a drug prescription database. Completeness of ascertainment was estimated with traditional methods based on the hypergeometric distribution and with a log-linear model.
Identification sources were not independent because the drug prescription database was positively related to family physicians and negatively related to the diabetes center (P < 0.001). Thus, completeness of ascertainment was overestimated (87.5% [95% CI 86.3-88.8]) when using only family physicians and the drug prescription database and underestimated (45.9% [43.9-48.1]) when using only the diabetes center and the drug prescription database. Because of characteristics contributing to variable "catchability" (probability of ascertainment), the estimated proportion of ascertainment increased with increasing time since diagnosis from 65.6% in the first tertile (<6 years) to 91.5% in the third tertile (>12 years); moreover, the ascertainment was estimated to be nearly complete (97.9%) for insulin-treated patients and scanty (28.9%) for diet-treated patients.
Because identification sources are likely to be dependent, the capture-recapture method should be used with caution in diabetes epidemiology and possibly when at least 3 sources are available. The subgroup of diabetic patients whose existence is inferred by this technique likely consists of newly diagnosed patients with mild disease severity.
本研究利用维罗纳糖尿病研究的数据，验证捕获 - 再捕获方法的一个主要假设（来源独立性），并描述所有来源均遗漏的已知糖尿病患者亚组的特征，该亚组患者数量通过捕获 - 再捕获方法进行估计。
维罗纳糖尿病研究于1986年12月31日通过三种不同来源确定了7148例2型糖尿病患者：家庭医生、糖尿病中心和药物处方数据库。基于超几何分布的传统方法和对数线性模型对确定的完整性进行了估计。
识别来源并非独立，因为药物处方数据库与家庭医生呈正相关，与糖尿病中心呈负相关（P < 0.001）。因此，仅使用家庭医生和药物处方数据库时，确定的完整性被高估（87.5% [95% CI 86.3 - 88.8]），而仅使用糖尿病中心和药物处方数据库时，确定的完整性被低估（45.9% [43.9 - 48.1]）。由于导致“可捕获性”（确定概率）变化的特征，确定的估计比例随着诊断后时间的增加而增加，从第一个三分位数（<6年）的65.6%增加到第三个三分位数（>12年）的91.5%；此外，胰岛素治疗患者的确定估计几乎是完整的（97.9%），而饮食治疗患者的确定情况很少（28.9%）。
由于识别来源可能相互依赖，在糖尿病流行病学中以及可能在至少有三个来源可用时，应谨慎使用捕获 - 再捕获方法。通过该技术推断存在的糖尿病患者亚组可能由疾病严重程度较轻的新诊断患者组成。
